otti0815 (@Otti@chaos.social) (@otti0815) header
otti0815 (@Otti@chaos.social) (@otti0815) profile picture

Doing something with clouds that machines can learn stuff faster

Munich
Joined September 2008
Photo by otti0815 (@Otti@chaos.social) (@otti0815)Photo by otti0815 (@Otti@chaos.social) (@otti0815)Photo by otti0815 (@Otti@chaos.social) (@otti0815)Photo by otti0815 (@Otti@chaos.social) (@otti0815)Photo by otti0815 (@Otti@chaos.social) (@otti0815)Photo by otti0815 (@Otti@chaos.social) (@otti0815)Photo by otti0815 (@Otti@chaos.social) (@otti0815)Photo by otti0815 (@Otti@chaos.social) (@otti0815)Photo by otti0815 (@Otti@chaos.social) (@otti0815)Photo by otti0815 (@Otti@chaos.social) (@otti0815)
‼️ BREAKING: OVHcloud's official CLI was hijacked in a supply chain attack. A hacker used a developer's GitHub account to plant a binary built to grab secrets in versions v0.11.0 to v0.15.0.
33
110
20
768
70,465
Finally, Apple has reached feature parity with GNOME
Well, bugOS 27 is unreal.
92
1,290
32
28,171
1,552,631
Run Kimi K3 on a Mac Studio 🫰 K3 is 2.8T parameters and 1.6TB on disk, which makes it impossible to run on Apple Silicon. Until now. Our MLX port is now open source: github.com/PipeNetwork/kimi-… To accomplish this, we solved two things: 1. We wrote a streaming converter that walks one layer at a time, so that mlx_lm doesn't need to materialize the whole model. 2. REAP pruning sits on top and scores all 896 experts against a calibration corpus to keep only ones your workload needs. That's what brings K3 down to 350GB and inside a Mac Studio.
Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale. Model weights: huggingface.co/moonshotai/Ki… Tech report: github.com/MoonshotAI/Kimi-K… Tech blog: kimi.com/blog/kimi-k3
122
316
136
2,962
1,360,169
Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale. Model weights: huggingface.co/moonshotai/Ki… Tech report: github.com/MoonshotAI/Kimi-K… Tech blog: kimi.com/blog/kimi-k3
1,539
7,162
2,352
45,885
14,756,774
Es ist mal wieder so weit das #Kulturspektakel in #Gauting findet statt und #FFMUC ist wieder gut vertreten. kulturspektakel.de/ map.ffmuc.net/#/en/map/f6bf6…
2
4
199
DLSS 5 is just a reverse TikTok Filter for games
48
What does a typical day on RIPE Atlas look like? Internet researchers analysed 24 hours of measurements - nearly 200,000 experiments and billions of results - to explore how RIPE Atlas operates at scale. 👉 Read on RIPE Labs: labs.ripe.net/author/yevheni… #RIPEAtlas
1
4
5
352
Wenn das Underlay stolpert, rettet das Overlay. Darin sind wir Experten ;). Und die MTU sagt leise „nein“. #meshthings #Freifunk #VPN #Telekom teapawt.org/telekom_hilft/status/2…
Replying to @Callsign_Smokey
Hi, es kann mit einer VPN-Verbindung funktionieren. Grüße ^Sarah
5
133
10,595
Ok? @OpenDNS @OpenDNSHelp listed our Public Resolvers as Proxy/Anonymizer. It's not working for some in the @lufthansa @MUC_Airport Wifi. Please fix this! doh.ffmuc.net dot.ffmuc.net doq.ffmuc.net More info: ffmuc.net/freifunk/dns/priva…
9
28
2,565
Clawdbot has the same hype as Rabbit R1. it was all over my feed back then called a brilliant product. game changer thing Clawdbot feels very familiar right now.
177
18
10
499
149,295
Built a little macOS menu bar app today called WhyFi. To answer the age old question: why is my Wi-Fi connection absolutely rubbish and/or not working right now? Personally I've wanted this for years! If people like it I'll package it up and make it available as a download.
118
108
27
3,794
294,095
We encourage all companies to follow the recent examples of @AlibabaGroup and @TencentGlobal and submit hand-written assembly patches (or other patches) to FFmpeg.
Hand written RISC-V assembly code written by @AlibabaGroup Cloud submitted to FFmpeg Up to 14 times faster than C. It's great to see so many corporate contributors of hand written assembly, a field historically dominated by volunteers!
8
31
6
741
35,986
bro casually walks and explains 5 GPU performance optimization methods for LLMs. one of the most simple and intuitive explanations for beginners.
170
2,010
121
17,388
946,849
German Tech having an insane resurgence in 2025 is one of the most underrated stories in European Tech. The country has had a fantastic 2025: > @traderepublic secured a €12.5bn valuation > @HelsingAI raised €600m at a €12bn valuation > @BlackForestLabs raised at a $3bn valuation > @n8n_io got valued at $2.5bn > @quantumdrones got valued at $3bn > @proximafusion raised a €130m Series A > @isaraerospace had its first launch > @MarvelFusion raised a €113m Series B extension > @ARXrobotics raised a Series A €42m These companies operate across AI, Defence, Robotics, Space, Energy, and Fintech; Germany is out-performing most of Europe when it comes to building hard technologies. These companies are also spread out across Germany. In France startups are concentrated in Paris, in the UK it's London, in Sweden it's Stockholm. However, what's remarkable about German tech is how spread out it is. Sure, Munich (which is the map) and Berlin are two big hubs, but Black Forest Labs is building in Freiburg. German tech is having a serious resurgence and it is GREAT to see it
47
145
16
1,259
90,891
Babe, wake up: new Nx feature just dropped! Thanks to @tvlabs I was able to put a 1 week spike on getting Nx.runtime_call through the finishing line. We can now interleave regular Elixir code in the middle of Nx computations. Works on EXLA compiled for CPU now, GPU coming soon.
10
2
62
12,715
#bsidesmunich2025 about to start! Looking forward to see a lot of you all in a minute
1
55
And I need to find a Sega Master System
12